AgentSkillsCN

Web Search Strategy

开展结构化研究，整合多方交叉验证的成果，并辅以置信度评分，提升研究结论的可靠性。

SKILL.md
--- frontmatter
name: Web Search Strategy
description: Conduct structured research with cross-referenced findings and confidence scoring
category: research
version: 1.0.0
triggers:
  - research-request
  - technology-evaluation
  - documentation-gap
globs: "**/docs/**,**/research/**"

Web Search Strategy Skill

Conduct structured research with cross-referenced findings and confidence scoring.

Trigger Conditions

  • •Research request for technology evaluation
  • •Need to validate technical decisions against current best practices
  • •User invokes with "research" or "technology evaluation"

Input Contract

  • •Required: Research question or topic
  • •Optional: Constraints (technology stack, budget, timeline)

Output Contract

  • •Ranked findings with confidence levels (high/medium/low)
  • •Cross-referenced sources (minimum 2 per finding)
  • •Recommendation with rationale

Tool Permissions

  • •Read: Project context for relevance filtering
  • •Web: Search and fetch capabilities
  • •Write: Research reports

Execution Steps

  1. •Decompose research question into specific search queries
  2. •Execute searches across multiple authoritative sources
  3. •Cross-reference findings (minimum 2 sources per claim)
  4. •Assign confidence levels based on source authority and agreement
  5. •Synthesize findings into structured report
  6. •Provide recommendation with caveats

Success Criteria

  • •Every finding cross-referenced from 2+ sources
  • •Confidence levels assigned to all claims
  • •Recommendation includes trade-offs and caveats
  • •Sources are authoritative and current (within 1 year)

Escalation Rules

  • •Escalate if findings are contradictory across authoritative sources
  • •Escalate if confidence is Low for a critical decision input

Example Invocations

Input: "Compare Temporal vs Restate for workflow orchestration in Go"

Output: Research report: Temporal (mature, large community, complex deployment) vs Restate (simpler deployment, newer, smaller community). Confidence: HIGH on feature comparison, MEDIUM on production readiness (Restate is newer). Recommendation: Temporal for production workloads requiring proven reliability; Restate for simpler use cases valuing operational simplicity.